Hollow-core fiber (HCF) has been attracting broad interests in recent decades, and has extended the communication window to longer wavelength, 2 micron. In this paper, we present a demonstration of low-latency HCF short-reach optical interconnection at 2 micron, achieving a high single-lane speed of 100 Gbps. High physical speed (near vacuum-light-speed) and high information speed (100-Gbps) optical interconnection with 100-m transmission distance is experimentally achieved with error-free performance. Probabilistically shaped discrete multi-tone (PS-DMT) modulation with entropy loading is employed to accommodate the deteriorated frequency response caused by the mode dispersion in HCF. Compared with conventional solid-core fiber (SCF), HCF contributes 30.95% (1.5-'s per kilometer) latency reduction for the physical link. The system level latency reduction is 28% for 1-km interconnection distance (11% for 100-m), with respect to a microsecond-scale switching network for typical datacenter applications. This work substantially leaps over the 100G milestone of 2-micron HCF optical interconnection, and shows the potentiality with lower latency and further higher capacity.

Multi-dimensional carrierless amplitude phase (CAP) modulation attracts a lot interest in recent optical communications as an advanced modulation format with improved performance. In this paper, we have demonstrated a multiband three-dimensional CAP (3D-CAP) modulation scheme and investigated its applications in short reach optical communication systems. Corresponding digital filters are designed by minimax algorithm with additional linear constraint to avoid the coding intersymbol interference. The performance of the digital filters is analyzed for optical fiber links with data rate over 100 Gb/s based on simulations. Similar to discrete multitone technique, several independent bands can be loaded with different 3D-CAP signals in the proposed multiband 3D-CAP system, offering improved tolerance to non-flat channels. 2-km single mode fiber transmission at total bit rate of 112.5 Gb/s is investigated through simulations and the results indicate better performance of multiband 3D-CAP compared with conventional 3D-CAP in short reach optical communication.

We demonstrated a support vector machine (SVM) based machine learning method to mitigate modulation nonlinearity distortion for PAM-4 and PAM-8 signals over a VCSEL-MMF optical link. Simulations at 100 Gbps data rate and experimental work at 60 Gbps data rate were carried out. We achieved a significant improvement in BER when complete binary tree SVMs (CBT-SVMs) are applied for both PAM-4 and PAM-8 signals. Quantitative analysis of the sensitivity gain versus modulation nonlinearity distortion is presented with experimentally verification. The results indicate that CBT-SVMs have better performance for PAM-8 compared to PAM-4. The sensitivity gain increases almost linearly with the increase of eye-linearity (increase of modulation nonlinearity distortion). Up to 2.5-dB sensitivity improvement is achieved by the proposed CBT-SVMs at eye-linearity of 1.72 for PAM-4.

Optical communications based on Nonlinear Fourier Transform (NFT) and digital coherent transceivers are proposed as a new theoretical framework for communications over the nonlinear optical fiber channel. For discrete eigenvalue transmissions (or soliton transmissions), one seeks to encode as much information as possible in each degree of freedom and shorten the distance between neighboring pulses to increase the overall bit rate. However, such attempts would result in nonlinear inter-symbol interference (ISI) across multiple symbols and significantly degrade transmission performance. In this paper, we investigated joint modulation of discrete eigenvalue λ and b -coefficents b (λ) and developed a suite of multi-symbol digital signal processing (DSP) techniques to exploit the statistical correlations between the continuous and discrete eigenvalues and b -coefficents to mitigate nonlinear distortions and improve detection performance. This include 1) jointly modulating both λ and b (λ) of pairs of 1-solitons so that the mean value of λ for solitons with odd index is α + 1 i while it is - α + 1 i for solitons with even index. This is followed by decoding superimposed received waveforms as 2-solitons with twice the INFT processing time window; 2) linear minimum mean squared error (LMMSE) estimation filters to mitigate noise in discrete eigenvalue λ using continuous eigenvalue; 3) multi-symbol (MS) LMMSE filters to mitigate noise in b (λ) using discrete eigenvalue noise and 4) approximate the received signal distributions of λ and b (λ) as Gaussians with mean and covariance matrices obtained empirically from experiments followed by Maximum Likelihood (ML) detection for each symbol or multi-symbol (MS)-joint ML detection of 2-soliton signals. We jointly modulate λ with 16-QAM and b (λ) with 16-APSK and a record single-polarization discrete eigenvalue transmission of 64 Gb/s (net 54 Gb/s) over 1200 km is experimentally demonstrated with the proposed multi-symbol DSP algorithms.

Modulation nonlinearity can severely distort multilevel modulation, and signal processing to mitigate the distortion is highly desirable. In this paper, we demonstrated a machine learning method for adaptive detection of 4-level pulse amplitude modulation (PAM-4) signals modulated by silicon micro-ring modulator (Si-MRM). The very limited linear modulation range of Si-MRM leads to serious modulation nonlinearity distortion for high-level modulations like PAM-4 with the consideration of wavelength drift. Our approach is based on the support vector machine (SVM) method, which can learn the nonlinear distortion of Si-MRM during PAM-4 modulation. Thus, the detection can be made adaptive for PAM-4 signals with nonlinear levels and level dependent noise. The modulation nonlinearity distortion of PAM-4 has been characterized in terms of level deviation (LD) with respect to wavelength drift. Up to 2.7-dB receiver sensitivity gain is obtained at about 26% LD by using the proposed SVM machine learning method. The receiver sensitivity-float range can be squeezed to be within 0.3 dB even with up to 30% LD, which indicates a stable detection of PAM-4 signals along with wavelength drift. Up to 3.63-dB receiver sensitivity improvement has been experimentally achieved at 50 Gbps for PAM-4 signals modulated by a Si-MRM and after 2-km standard single mode fiber transmission. The stable operation of Si-MRM is very difficult and very important. The proof-of-concept results indicate the very promising capability of machine learning method for stable detection of PAM-4 signals modulated by Si-MRM, which is of great significance for practical application of Si-MRM in optical interconnection.